Add code to generate SNRs for abitrary sensitivity curves

[avivajpeyi]

Currently, we have some cached injection SNRs using LVK design, O1, O3 sensitivity
https://github.com/TeamCOMPAS/COMPAS/blob/dev/compas_python_utils/cosmic_integration/selection_effects.py

However, we have no code to create new SNRs datasets for new PSD curves (say the O4 curve, or the updated O3 curve -- the one used SimNoisePSDaLIGOMidHighSensitivityP1200087 is outdated).

[ilyamandel]

@avivajpeyi , @stegmaja -- is this something one of you is planning to work on?

[avivajpeyi]

Hey @ilyamandel -- i assume we'll need this for our surrogate LnL work (so yes, once we get to testing with real data).

[ilyamandel]

@avivajpeyi , ping on my question above. Perhaps chat with @SimonStevenson , who recently updated the python scripts to correctly use the specified sensitivity.

[ilyamandel]

@avivajpeyi , ping to check if you are still looking at this one.

[avivajpeyi]

Woops, sorry, I forgot about this! I'll have a look.

[avivajpeyi]

@ilyamandel had some offline comments:

We want the ability to:

• Recompute SNRs for an arbitrary PSD (useful for someone who might be interested in what can be detected with some new version of a NEMO or Einstein Telescope++ noise curve)
• Use existing tools to estimate sensitivities for existing science runs from injection campaigns (I think LVK folks have already created these, so we shouldn't reinvent the wheel)

[avivajpeyi]

1. SNR curves: We could use bilby's inbuilt methods..
2. Sensitivity curves:

• Option1: O4 curves have been released https://arxiv.org/abs/2508.10638. We could grab the tabular data of simulated CBC parameters plus detection significance from zenodo, load them and compute the fraction of injections recovered to get the sensitive volume or detection probability for a given population (similar to what we are currently doing)
• Option2: https://arxiv.org/abs/2307.02765, we can use Reed's semianalytic method to predict the search sensitivity by modeling the matched-filter SNR distribution in Gaussian noise. I think this would be easier to try out different detectors (like NEMO/ET).

We could also compute pdet using Gwpop? https://colmtalbot.github.io/gwpopulation/guide/selection.html